Managers’ Self-Serving Attribution Bias and Corporate Financial Policies

Feng Li

Accounting Department Stephen M. Ross School of Business

University of Michigan E-mail: feng@umich.edu

Phone: 734-936-2771

July 12, 2010

Abstract

The self-serving attribution bias (“SAB”) refers to individuals taking responsibility for good outcomes and blaming others for bad outcomes. Consistent with the existence of managerial SAB, I find that managers tend to use more first-person pronouns (relative to second- and third-person pronouns) in the Management Discussions and Analysis Section of the 10-K filings when firm performance is better. A consequence of SAB is overconfidence (i.e., overestimating the mean and underestimating the variance of future cash flows). Consistent with this argument, managers with more SAB are more likely to issue forward-looking statements and make earnings forecasts, the tone (e.g., positive versus negative) of their forward-looking discussions has smaller variation, and their earnings forecasts tend to be overly optimistic. Firms whose managers have more SAB have higher investment-cash flow sensitivity and experience more negative market reactions around acquisition announcements. These firms also tend to have higher leverage, rely more on long-term debt financing, are more likely to repurchase stocks, and are less likely to issue dividends. Collectively, the evidence suggests that managers have self-serving attribution bias and this bias has implications for corporate policies. I thank Brian Bushee, Ilia Dichev, Yu Huang, Kenneth Merkley, Stephen Salterio, Frank Yu, and participants of the University of Michigan Kapnick Conference for comments.

1

“Solid execution of our strategies and the absence of significant catastrophes contributed to our outstanding results in 2006. Around the world and across all of our business segments we are capitalizing on growth opportunities, using our business diversity and matrix management structure to respond quickly to customer needs.”

American International Group, Inc. 2006 Annual Report

“AIG reported that the continued severe credit market deterioration, particularly in mortgage-backed securities, and charges related to ongoing restructuring activities, contributed to a record net loss for the fourth quarter of $61.7 billion, or $22.95 per diluted share, compared to a 2007 fourth quarter net loss of $5.3 billion, or $2.08 per diluted share.”

American International Group, Inc. 2008 Annual Report

1. Introduction

A central part of behavioral economics deals with economic agents’ overconfidence

and its economic consequences.1 One key mechanism through which people become

overconfident is the existence of self-serving attribution bias (SAB), which refers to

individuals taking responsibility for successful outcomes but blaming circumstances or

other persons for unsuccessful outcomes (Gervais and Odean 2001, Barber and Odean

2002, Malmendier and Tate 2005, 2008, Billett and Qian 2008).

While there is pervasive evidence that individuals exhibit SAB in lab experiments or

surveys, little direct evidence based on empirical data exists on whether corporate

executives have SAB and whether managerial SAB has economic consequences. This

paper attempts to provide such evidence. To test for possible management SAB, I

examine the language in the Management Discussion and Analysis (MD&A) section of

firms’ 10-K filings. Under the regulations of the Securities and Exchange Commission,

1 Malmendier and Tate (2005) define overconfidence as the underestimate of variance and/or overestimate of mean, while Baker et al. (2007) label the underestimate of variance as overconfidence and the overestimate of mean as optimism. I follow Malmendier and Tate in this paper.

2

firms are required to provide a narrative explanation of firm operations in the MD&A.

The MD&A is mandatory and subject to auditor’s examination and thus provides an ideal

setting with minimum selection bias to test for management SAB. Examining the

language in the MD&A is also consistent with the approach used extensively in the social

psychology literature, which generally studies the language patterns of the experimental

subjects to identify SAB (Miller and Ross 1975).

I use the percentage of first-person pronouns relative to that of second- and third-

person pronouns in the MD&A to measure managers’ tendency to self-reference.

Examining more than 30,000 MD&As filed between 1994 and 2007 by U.S. public firms,

I find that consistent with managers having self-serving attribution bias, their self-

referencing tendency is positively associated with firm performance as measured using

stock returns during the fiscal year.

However, this result might be explained by unobserved management characteristics.

Firms might perform better because managers are more capable and more capable people

perhaps like to use more self-referencing words. Managers may also feel more confident

when firm performance is better and more confident people may like to use more first-

person pronouns. To mitigate these concerns, I first examine firms whose performance is

close to the average firm in the same industry. The contribution of managers’ ability to

these firms’ performance is relatively small. However, I find that managers’ self-

referencing tendency still positively relates to firm performance for these firms.

I then explore factors that should correlate with the degree of SAB. Prior research in

psychology has identified individuals’ cognitive limitations, including selective attention

and informational accessibility in memory, as one of the main explanations for SAB

3

(Miller and Ross 1975). Therefore, the noisier the performance feedback, the more severe

people’s self-serving attribution bias is. Cognitive theories also predict that SAB is an

increasing function of the ex ante expectation for success (Campbell and Sedikides 1999,

Shepperd et al. 2008) because when outcomes disconfirm (confirm) a positive

expectation, people tend (not) to initiate a search for the cause of the failure. Consistent

with these predictions, I find that managerial SAB increases with firm size, the number of

business segments, and past performance.

To further mitigate the concerns about omitted variables, I explore within-MD&A

sentence-level variation. I examine the attribution process reflected in the sentences that

contain causative words (e.g., “because”). The sentence-level analysis can control for any

unobserved firm-year specific factor by allowing interactions of firm and year fixed

effects in the regression. I find that within an MD&A, managers use more self-

referencing pronouns in the causative sentences with more positive tone. This evidence

shows that unobserved firm-year specific factors (such as management ability or

confidence) are unlikely to explain the association between performance and managers’

self-referencing tendency.

Next, I link management SAB to management overconfidence. If SAB is one

mechanism through which managers become overconfident, then the degree of SAB is

expected to positively relate to the extent of overconfidence, i.e., underestimating the

variance and overestimating the mean of future cash flows. I classify managers as having

SAB if the MD&A sections contain abnormally high percentage of self-referencing. I

find that managers with SAB are more likely to issue forward-looking statements in

MD&As and make earnings forecasts. Use a Bayesian machine learning algorithm to

4

measure the positive/negative tone of the forward-looking sentences in MD&As, I find

that the tone of these sentences made by managers with SAB tends to be more

homogeneous, indicating that they possess a more precise belief about their firms’ future

cash flows. The earnings forecasts made by managers with SAB are also overly

optimistic. Collectively, these results suggest that SAB leads to overconfidence.

Finally, I examine the implications of managerial SAB for investment and financing

policies. Prior research documents that overconfident managers tend to overinvest, make

poor acquisitions, rely more on debt financing, are more likely to repurchase stock, and

are less likely to issue dividends (Malmendier and Tate 2005, 2008, Ben-David et al.

2007, Billet and Qian 2008, Malmendier et al. 2010). If SAB leads to overconfidence,

similar behavior is expected for managers with SAB. The empirical results are consistent

with this prediction. I find that firms whose managers have SAB have higher investment-

cash flow sensitivity, more negative market reactions to acquisition announcements,

higher leverage, higher proportion of long-term debt, higher probability of repurchasing

stocks, and lower probability of issuing dividends.

The results in this paper provide direct support for the existence of managerial self-

serving attribution bias and show that the individual SAB documented extensively in lab

experiments and surveys in the psychology literature (Miller and Ross 1975, Sedkides et

al. 1998) is also observed in corporate managers. This study also contributes to the

growing behavioral corporate finance literature (see Baker et al. 2007 for a review) by

showing that self-serving attribution bias is a cognitive reason for management

overconfidence. Furthermore, using textual analysis to measure managers’ behavioral

bias provides a methodology for future research on management SAB.

5

The rest of the paper proceeds as follows. Section 2 describes the data and empirical

measures. Section 3 presents the empirical evidence. Section 4 discusses the implications

of the empirical results and Section 5 concludes.

2. Data and empirical measure of managers’ self-references

I construct my sample based on the intersection of firm/years available on the

EDGAR filings database maintained by the Securities and Exchange Commission (SEC)

and the Compustat annual file for years 1994-2007. These databases are manually joined

based on Compustat GVKEY and the SEC’s Central Index Key. Firms without matches

are dropped from the sample. For each firm-year observation I download the

corresponding 10-K filing. Filings with less than 3,000 words or 100 lines are dropped. In

addition, I exclude firms without CRSP data to compute stock returns. I extract the

MD&A section of the 10-K filings following the process employed in Li (2010). I require

that each MD&A contain a minimum of five hundred words to ensure that the empirical

measures derived from it are not due to random noise. These procedures result in a

sample of 38,953 firm-year observations. Depending on the specific empirical tests and

control variables, the sample size varies.

I use the Linguistic Inquiry and Word Count (LIWC) package to compute the self-

referencing measure. LIWC is a text analysis software program designed by

psychologists James W. Pennebaker, Roger J. Booth, and Martha E. Francis; the program

is able to calculate the degree to which people use different categories of words across a

6

wide array of texts.2 The default LIWC dictionary consists of 2,300 words and word

stems with each word or word stem defining one or more word categories.

I compute managers’ self-referencing tendency as the percentage of self-referential

pronouns relative to the pronouns that refer to other parties in the MD&A:

,OtherWeSelf −=

where We is the percentage of first-person pronouns (20 words in the LIWC

dictionary such as “I,” “We,” and “Our”), and Other is the sum of the percentages of

second-person pronouns (14 words in the dictionary such as “You” and “Your”) and

third-person pronouns (22 words in the dictionary such as “She,” “Their,” and “Them”).

Table 1 shows the descriptive statistics of these measures and other variables used in

the empirical tests. The mean values of We and Other are 1.27 percent and 0.12 percent,

respectively. For comparison, Panel B of Table 1 presents the mean of these two

variables for other text sources documented in Pennebaker et al. (2001, Table 3). On

average, “emotion writing” has 3.5 percent self-referring pronouns and 2.1 percent other-

referring pronouns; books have 0.5 percent self-referring pronouns and 1.3 percent other-

referring pronouns. Therefore, the frequency of first-person pronouns of MD&A text is

between “emotion writing” and “books.” However, the frequency of second- and third-

person pronouns is much lower in the MD&A setting. This is likely because managers

seldom use words like “you” or “they” in the MD&A; rather, they tend to use generic

words such as “the economy” to discuss external factors. The empirical results in this

paper are based on Self; however, all empirical inferences remain the same if We is used

to measure managers’ self-referring tendency.

2 More details about the software can be found at http://www.liwc.net/ and http://homepage.psy.utexas.edu/homepage/Faculty/Pennebaker/Home2000/Words.html.

7

The change in managers’ self-referencing tendency, ΔSelf, is calculated as the year-

to-year change in Self. Table 1 shows that there is substantial variation in ΔSelf: the

standard deviation of ΔSelf is 0.86 and the inter-quarter range is 0.19, which are

significant compared with the mean value of Self (1.15).

3. Empirical results

3.1 Managers’ self-referencing tendency and firm performance

Table 2 presents the regression results of ΔSelf on firm performance and other control

variables. Examining ΔSelf, rather than Self, offers several advantages. First, it helps filter

out time-invariant factors that might drive both firm performance and managers’ self-

referencing tendency. Second, a change specification can remove the stale information

because prior literature has shown that there is substantial boilerplate in MD&A that does

not change over time (Brown and Tucker 2010). I measure firm performance using the

twelve-month stock returns during the fiscal year. Year and industry fixed effects are

included in the regressions. Since a firm’s auditor is involved in the preparation of the

firm’s 10-K filing, auditor fixed effects are also included.

Column (1) in Table 2 presents the univariate regression of ΔSelf on the stock returns

(Ret). The t-statistics are calculated based on standard errors that are clustered at the year

level. The positive coefficient on Ret (0.11 with a t-statistic of 4.40) suggests that

managers have self-serving attribution bias. The economic magnitude of the coefficient is

significant. For instance, the inter-quartile range of Ret is 0.61, implying that an inter-

quarter change in Ret impacts the dependent variable (ΔSelf) by 0.07 (calculated as 0.61 *

0.11), which is about one-third of the inter-quartile range of ΔSelf (Table 1).

8

In column (2), I control for several firm characteristics, including Tobin’s Q, log of

book value of assets, and firm age. Q is the market value of the assets divided by the

book value at the end of the fiscal year. Firm age is the average number of years since a

firm shows up in CRSP monthly stock return file. The coefficient on Ret remains positive

(0.09) and statistically significant (t-statistic = 4.98).

However, unobserved time-varying management ability might explain the positive

association between performance and managers’ self-referencing tendency. A firm might

perform better because its managers are more capable compared with last year. If this is

true, then to the extent that more capable people tend to use more self-referencing words,

a positive association between performance and the frequency of self-referencing words

in MD&As is observed. Another possible explanation is that when firm performance is

better, managers tend to be more confident and more confident people may like to use

more self-referencing words.

To mitigate these concerns, I examine two sub-samples. The first sample is a set of

firms whose performance is within a relatively small range of that of the average firm in

the same industry. The second sample consists of firms with small idiosyncratic stock

return volatility, i.e., the stock returns of these firms can be explained mostly by the

market and industry returns. The performance of firms in these two sub-samples closely

mimics that of the market and hence the contribution of managers’ ability to performance

is relatively small. If there is evidence of SAB for these two sub-samples, then it is less

likely that the result is driven by the unobserved management ability.

Column (3) in Table 2 presents the results for firm-years whose stock returns are

between the 25th percentile and the 75th percentile of the returns of the firms in the same

9

two-digit SIC industry in the same year. To make sure that the industry definition is

meaningful, I require that each industry has at least ten firms in the sample for a given

year. The results indicate that Ret is still positively and significantly associated with ΔSelf.

In fact, the coefficient on Ret (0.14 with a t-statistic of 3.08) is larger that in column (2).

In column (4), the empirical tests are conducted using firm-years whose stock returns are

within the +/- 5% range of the median return of firms in the same industry. Again the

results show that there is positive relation between firm performance and the frequency of

self-referencing words. Column (5) shows the results based on firms in the bottom

quartile of idiosyncratic return volatility, which is calculated as the standard deviation of

the residuals from the firm-specific regression of monthly stock returns on the CRSP

value-weighted market index returns and the two-digit SIC industry returns using data

from the fiscal year. The results based on this subsample are similar to those in columns

(1) and (2): the coefficient on Ret is 0.08 with a t-statistic of 2.36.

3.2 Cross-sectional variations in the self-serving attribution bias

Examining cross-sectional variations based on the theoretical determinants of SAB

provides another approach to mitigate the concerns about omitted variables. The social

psychology literature argues that people have SAB because of their limited cognitive

ability due to selective attention, information availability, or accessibility in memory

(Miller and Ross 1975). Individuals often stop well short of examining all possible

explanations for an outcome, accepting instead the first logical explanation that comes to

mind, a search strategy that is satisfying and requires the least amount of effort (Simon

1956, Kunda 1990, Pyszczynski and Greenberg 1987). To the extent that a noisier

10

performance feedback is more likely to lead to selective information gathering, SAB is

expected to be more severe in a noisier information environment.

Another prediction of the cognitive explanations for SAB is that people’s ex ante

expectations for success is positively associated with the ex post attribution process

(Campbell and Sedikides 1999). This is because when outcomes confirm a positive

expectation, people do not search for an explanation for why the outcome occurred

(rather, they generally fall back on the explanation that gives rise to the positive

expectation; when outcomes disconfirm a positive expectation, people initiate a search

for the cause of the failure.

Based on these cognitive explanations for SAB, I explore three factors as cross-

sectional determinants of SAB. First, larger firms are more complex and their

performance feedback is likely to be noisier. In addition, managers of larger firms are

likely to be more successful in the past and therefore have higher expectations about their

performance. This reasoning suggests that managers of larger firms have more significant

SAB. I measure firm size using Size, log of the market value of equity at the end of a

fiscal year. Second, I expect that managerial SAB increases with the number of business

segment (Num_seg, log of the number of business segment reported in Compustat), a

proxy for the complexity of a firm’s information environment. Third, I expect managerial

SAB to increase with a firm’s past success, measured using a dummy variable Success

that equals 1 if a firm’s stock return is above industry median in both of the last two years

and 0 otherwise.

Columns (1) to (3) in Table 3 show that larger firms, firms with more business

segments, and more successful firms are indeed more likely to have management SAB, as

11

indicated by the positive and significant coefficients on the interactions of Size, Num_seg,

and Success with Ret. In column (4), the three factors are included together in the

regression and the results show that all three interaction terms are still positive and

statistically significant. Overall, the evidence supports the predictions of the cognitive

explanations that management SAB is an increasing function of information noisiness

and expectation for success.

3.3 Managers’ self-referencing tendency in causative sentences

Another way to mitigate the concern about omitted firm-year specific factors is to

examine the sentences in the MD&As. I focus on sentences in MD&As that contain

causative words, such as “because” and “hence” (“causative sentences”). By definition,

managers use these sentences to explain something. If managers have SAB, then they are

more likely to use self-referencing words in causative sentences which have positive tone,

i.e., when they express something positive, the managers are more likely to attribute it to

themselves. A significant advantage of the sentence-level analysis is that it allows the

interactions of firm and year fixed effects in the regressions to control for firm-year

specific factors.

To implement this test, I first extract all the sentences in the MD&As that contain at

least one of the 49 causative words based on LIWC. This procedure yields about 3.5

million causative sentences. Following Li (2010), I measure the tone of the causative

sentences using a Naïve Bayesian machine learning algorithm (Manning and Schutze

1999).3 For every sentence, the Naïve Bayesian algorithm predicts the probability of the

3 To implement the Naïve Bayesian learning algorithm, I first manually categorize 30,000 sentences of randomly-selected statements extracted from corporate MD&As into three different tones (e.g., positive,

12

sentence having positive, neutral, or negative tone. I define a dummy variable

Positive_tone for each sentence, which equals one if the probability of the sentence

having positive tone is the highest among the three categories as predicted by the Naïve

Bayesian algorithm and zero otherwise.

I then regress Self, the percentage of the first-person pronouns minus the percentages

of second- and third-person pronouns of the sentence, on Positive_tone and control

variables. In column (1) of Table 4, firm, year, and auditor fixed effects are included, the

coefficients on Positive_tone is 0.32 (t-statistic clustered at year level = 4.78). In column

(2), the interactions of firm and year fixed effects are included in the regression; since

there are about 8,000 firm fixed effects and 13 year fixed effects, this translates into more

than 10,000 (unreported) fixed effect coefficients in the regression. The results in column

(2) show that in the causative sentences with positive tone, managers are more likely to

use self-referencing words compared with those with negative tone in the same MD&A

(the coefficient on Positive_tone is 0.37 with a t-statistic clustered at year level of 4.68).

The coefficients in column (2) are estimated based on variations within a specific MD&A.

Therefore, any firm-year specific factors, such as the self-confidence of managers, do not

explain the finding in this column. In columns (3) and (4), the same empirical tests as

those in columns (1) and (2) are conducted using firms with at least 8 years of data in the

sample and the results are similar.

In summary, this evidence based on the causative sentences suggests that the

unobserved firm-year specific variables (such as management ability or confidence) are

neutral, and negative tone). These manually-coded sentences are then used as training data in a Naïve Bayesian machine learning algorithm to classify the tone of other forward-looking statements. Details of the implementation follow Li (2010).

13

unlikely to explain the positive association between firm performance and managers’

self-referencing tendency.

3.4 Self-serving attribution bias and managerial overconfidence

Next, I examine the link between management SAB and overconfidence. SAB is a

key factor that fosters overconfidence. Even if individuals start out without

overconfidence, the attribution bias can lead them to become overconfident, as

demonstrated in Gervais and Odean (2001). Linking management SAB to managers’

overconfidence also lends further support to the argument that the positive relation

between firm performance and managers’ self-referencing tendency documented in

Tables 2 to 4 captures managers’ self-serving attribution bias.

I estimate managerial SAB as the abnormal level of self-referencing using the

following regression by pooling all firm-years together:

ititit tSelf εβα ++= Re .

The residual itε̂ is then calculated as )Reˆˆ(ˆ ititit tSelf βαε +−= , where α̂ and β̂ are

the estimated intercept and slope in the regression. itε̂ captures the amount of managerial

self-referencing that is not explained by the average level of self-serving attribution bias

and firm performance. To capture managers’ SAB, I construct a dummy variable SABit

that equals 1 if itε is greater than the median value of my sample firm-years and 0

otherwise.

Overconfidence implies overestimating the mean and underestimating the variance of

future cash flows (Malmendier and Tate 2005). I link SAB to overconfidence using two

settings that reflect managers’ belief about future cash flows: the forward-looking

14

statements in the MD&As and earnings forecasts. First, I examine the association

between SAB and manager’s tendency to issue forward-looking statements and provide

earnings forecasts. If managers underestimate the variance of future cash flows, they are

more likely to issue forward-looking statements and earnings forecasts.4 To test this

implication, I extract all the forward-looking statements from the MD&As using the

algorithm in Li (2010). I also obtain the forecasts of next quarter’s earnings per share

made by managers from the Thomson Reuters First Call Company Issued Guidelines

database.

Table 1 Panel A shows that the mean value of NFLS, the number of forward-looking

sentences in MD&As, is 228. In the empirical analysis, I focus on its log value,

Ln(NFLS), because of the skewness in the variable, but the empirical inferences based on

the raw values remain the same. The mean value of Forecast, a dummy variable which

equals one if a manager issues forecasts of next quarter’s earnings afrter the 10-K filing

date, is 0.11.

Column (1) in Table 5 presents the regression results of Ln(NFLS) on SAB with firm

size, Q, firm age, current earnings, stock return volatility, and year and industry fixed

effects as control variables. Managers with more SAB appear to make more forward-

looking statements in their MD&As, as indicated by the positive and significant

coefficient on SAB (0.049 with a t-statistic of 7.91). Given that the standard deviation of

Ln(NFLS) is 0.73 (Table 1), the result suggests that SAB explains about 7 percent of the

variation in Ln(NFLS). The results are also robust to including firm fixed effects instead

of industry fixed effects (column (2)).

4 The earnings forecasts by managers have significant implications for stock prices (Patell 1976, Penman 1980, Nagar et al. 2003, Hutton and Stocken 2007) and providing good earnings forecasts by managers can potentially reduce the cost of capital (Coller and Yohn 1997).

15

Columns (3) and (4) estimate the logistic regression of Forecast on SAB and control

variables. The coefficient on SAB is 0.103 (z-statistic = 2.87) in column (3), indicating

that managers’ self-serving attribution bias is positively associated with the likelihood of

managers issuing earnings forecasts. Unreported results show that the marginal effect of

SAB is about 1 percent. Given that on average about 11 percent of managers make

earnings forecasts, this effect is economically significant. A similar result obtains in

column (4) when firm fixed effects are included.

I then directly examine the mean and variance of managers’ expectation about future

cash flows. I calculate earnings forecast error, FERR, as earnings per share forecasts

minus actual earnings per share divided by book value of assets per share. A positive

FERR indicates that managers overestimate the mean of future cash flows. Column (5) in

Table 5 shows the regression result of FERR on SAB with year and industry fixed effects

included. The coefficient on SAB is 0.052 with a t-statistic of 3.15. This suggests that

managers with self-serving attribution bias have overly optimistic estimates of future

cash flows. Since the standard deviation of FERR is 0.55 (Table 1), the result suggests

that SAB can account for about 10 percent of the standard deviation of FERR. The results

in column (6) are similar after controlling for firm fixed effects.

To measure the variance of managers’ belief about future performance, I compute the

variation in the tone of the forward-looking statements in MD&As. For each MD&A with

at least three forward-looking statements, I calculate the index of qualitative variation

(Healy 2008) as:

2/)](1[3 21,

20

21 −++−×= itititit PPPIQV ,

16

where IQVit is the index of qualitative variation for firm i in year t and Pit1 (Pit0, Pit,-1)

is the percentage of sentences with positive (neutral, negative) tone. The index measures

the dispersion of the distribution of a qualitative variable; it is bounded between 0 and 1

and is increasing in the dispersion. If one category dominates (e.g., all sentences are

positive and there is no variation in tone), then the index has a value of zero. If each

category is represented equally, then the index equals 1.

The mean value of IQV is 0.81 (Table 1), suggesting that the tone of the forward-

looking statements within an MD&A has substantial variations. Column (7) in Table 5

shows that SAB is negatively associated with IQV (the coefficient on SAB is -0.003 with a

t-statistic of -1.92), suggesting that managers with more SAB tend to issue forward-

looking statements in the MD&A that are more homogeneous in tone. In columns (8),

firm fixed effects are included and the coefficient on SAB is -0.007 (t=-4.32). Given that

the standard deviation of IQV is 0.14 (Table 1), this result suggests that managerial self-

attribution bias can explain about 5 percent of the standard deviation of IQV.

3.5 Self-serving attribution bias and investment-cash flow sensitivity

Overconfident managers systematically overestimate the return to their investment

projects or underestimate the variance of future cash flows and hence use lower discount

rate in valuation (Malmendier and Tate 2005; Ben-David et al. 2007). With sufficient

internal funds for investment, these managers overinvest relative to the first-best. Without

sufficient internal funds, however, they are reluctant to raise capital through external

financing because they perceive the stock of their company to be undervalued by the

17

market and hence reduce the investment. If self-serving attribution bias leads to

overconfidence, a positive association between SAB and overinvestment is expected.

I estimate the relation between SAB and the investment cash-flow sensitivity in Table

6, where the sensitivity of investment to cash flows is a proxy for overinvestment

(Fazzari et al. 1988). The dependent variable is the amount of capital expenditure scaled

by the beginning value of Property, Plant, and Equipment. The independent variables

include cash flows and its interaction with SAB and other control variables. Column (1)

presents the baseline results without any control variable. The coefficient on the

interaction of SAB with cash flow is positive (0.223) and statistically significant. The

result is robust to clustering the standard errors by firm, including the interactions of

industry fixed effects and cash flows, including the interactions of year fixed effects and

cash flows, and including firm fixed effects (columns 2 to 6). The evidence shows that

managers who demonstrate a higher level of SAB tend to overinvest. The economic

magnitude is comparable to that documented in Malmendier and Tate (2005). For

instance, Malmendier and Tate (2005) document that overconfident CEOs are associated

with an investment-cash flow sensitivity higher by 0.2339 (column V in Table 5 of their

paper). In column (5) the coefficient on SAB * CF is 0.225, suggesting that the

investment-cash flow sensitivity is higher by 0.225 when managers have SAB.

3.6 Self-serving attribution bias and acquisition announcement returns

Overconfident managers tend to overestimate their ability to obtain synergies from a

merger and as a result they overpay for target companies and undertake value-destroying

mergers (Malmendier and Tate 2008, Billett and Qian 2008). To the extent that self-

18

serving attribution bias is a factor that leads to overconfidence, I test whether managers

with SAB are more likely to make value-destroying acquisitions.

I use the SDC Platinum database to identify all completed acquisitions by my sample

firms of private, public, and subsidiary targets. Following Moeller, Schlingemann, and

Stulz (2004) I select M&A deals that satisfy the following criteria: (1) the target firm has

to be a U.S. firm; (2) the deal value is at least $1million; (3) the percentage stake that the

acquirer seeks in the deal is at least 50%; and (4) the deal is completed. This procedure

yields a sample of 7,712 acquisitions. CAR[-1,1], the cumulative abnormal return in the

three days around the deal announcement date, is then calculated where day 0 is the deal

announcement date.

Table 7 presents the regression results of CAR[-1,1] on SAB and other control

variables with different specifications. Firms with managers who have SAB experience

significantly more negative reactions to merger announcements. For instance, column (3)

indicates that the announcement return to mergers done by managers with SAB is lower

by 0.5 percent, after common firm and deal characteristics and year and firm fixed effects

are controlled for. While this effect is slightly smaller than the effect documented by

Malmendier and Tate (2008), who find that mergers done by overconfident managers

experience a market reaction lower by 0.8 percent, it is still substantial.

In summary, the evidence based on the investment-cash flow sensitivity and the

acquisition announcement return suggests that managers with SAB are more likely to

overinvest and invest sub-optimally.

3.7 Self-serving attribution bias and corporate financing decisions

19

Malmendier et al. (2010) and Ben-David et al. (2007) show that because

overconfident managers view equity financing as costly, they tend to use higher leverage,

rely more on long-term debt, are more likely to repurchase shares, and are less likely to

issue dividends. In this section, I examine whether managers with SAB make similar

financing decisions.

Column (1) in Table 8 shows that management SAB is positively associated with a

firm’s leverage (defined as the book value of debt divided by the sum of book value of

debt and market value of equity) after year and industry fixed effects and other control

variables are included in the regression. When managers have SAB, the leverage is

higher by 0.038. Given that the standard deviation of firm leverage is 0.26 (Table 1), this

shows that SAB explains more than 10 percent of the standard deviation of firm leverage.

The effect is smaller but still economically and statistically significant when firm fixed

effects are controlled for in column (2).

Columns (3) and (4) focus on the percentage of long-term debt (relative to the total

amount of debt). Ben-David et al. (2007) show a positive and significant relation between

management overconfidence and the percentage of long-term debt. Consistent with the

association between SAB and overconfidence, results in columns (3) and (4) show that

managers with SAB also tend to rely on more long-term debt financing.

Columns (5) and (6) report the logistic regressions of Repurchase, a dummy variable

that equals 1 if purchase of common and preferred stock by a firm in a given year is

greater than 1 percent of equity and zero otherwise, on SAB and other control variables.

The coefficient on SAB is 0.069 in column (5) with year and industry fixed effects

included and 0.082 in column (6) when year and firm fixed effects are controlled for. In

20

both cases, the effect is statistically significant. Unreported marginal effect calculation

shows that a manager with SAB is 1.5 percent more likely to repurchase stocks.

Column (7) shows a negative relation between SAB and Dividends, a dummy variable

that equals 1 if a firm declares dividends in a year and zero otherwise, in a logistic

regression setting. Since dividend policy tends to be sticky, firm fixed effects cannot be

included in the analysis. The coefficient on SAB is -0.076 (z-statistic=-2.24). Unreported

result shows that the marginal effect of SAB is -1 percent, indicating that the probability

of a firm issuing dividends when the managers have SAB is lower by 1 percent.

Overall, the financing policies by mangers with SAB are consistent with the

hypothesis that these managers view equity as undervalued and as a result prefer debt

financing, repurchase stocks more often, and avoid paying dividends.

4. Discussions

4.1 Existence of management self-serving attribution bias

The behavioral finance literature shows that irrational market participants can stay in

the market for a long time without being forced out by rational agents (De Long et al.

1991, Odean 1998, Daniel et al. 1998, Gervais and Odean 2001). The literature also

examines the implications and consequences of overconfident participants (Malmendier

and Tate 2005, 2008, Malmendier et al. 2010, Ben-David et al. 2007).

A key assumption in the overconfidence models is the economic agents’ self-serving

attribution bias. For instance, Gervais and Odean (2001) model a trader’s overconfidence

in a dynamic learning model where the trader learns about her ability over time. A key

21

ingredient of this model is the trader’s self-serving attribution bias modeled as her

overweighing past performance success.

Given the role of the self-serving attribution bias in the overconfidence literature, it is

important to empirically test for the existence of such bias. While there is a substantial

literature documenting individuals’ SAB, the evidence is mainly from lab experiments or

surveys by psychologists (Larwood and Whittaket 1977, Russo and Schoemaker 1992,

Shepperd et al. 2008). One exception is Choi and Lou (2008) who measure the self-

serving attribution bias of investors indirectly using past performance volatility. In the

management literature, researchers use the frequency of first person singular pronouns

relative to first person plural pronouns as an indication of “narcissism” and study its

implications for firm practices (Chatterjee and Hambrick 2007). Compared with the prior

studies, this paper provides direct evidence on the self-serving attribution bias of

corporate managers.

4.2 Causes of SAB and implications for the management overconfidence literature

The social psychology literature has proposed two broad streams of explanations for

SAB: cognitive and motivational. As discussed in Section 3.2, the cognitive explanation

argues that it is the limited information-processing capacity of individuals that drives the

SAB (Miller and Ross 1975).

On the other hand, the motivational approach argues that people make internal

attributions for success and external attributions for failure to maintain their self-esteem

and feel good about themselves (Zuckerman 1979). Two distinct motives are proposed:

self-enhancement and self-presentation. The self-enhancing motivations help individuals

22

protect self-esteem by creating causal explanations that serve to make themselves feel

better. The self-presentation motivation refers to the drive to convey a desired image to

others (Schlenker 1980), i.e., although people may not believe the content of a self-

serving utterance, they may nevertheless offer it to others in order to create a favorable

impression.

The psychology literature has had a long debate on which of the two explanations is

the main factor that leads to the self-serving attribution bias. Miller and Ross (1975)

review the literature and point out that cognitive reasons could be sufficient in explaining

SAB. However, more recent research in social psychology has accumulated substantial

evidence that motivational reasons might be the dominant factor that drives SAB

(Sedikides et al. 1998, Sheppard et al. 2008).

Whether cognitive or motivational factors drive the SAB of market participants has

significant implications for the overconfidence research in behavioral economics. If

managers have SAB primarily to seek additional (self-enhancing or self-presenting)

motivations, then the SAB seems less likely to lead to overconfidence with real decision

consequences. However, the cognitive explanation is a more plausible factor with real-

consequence implications because a key difference between the cognitive and

motivational explanations is that economic agents with cognitive limitations s are less

likely to realize that they have the bias.

The empirical results in this paper show that the cognitive explanations are at least

partially responsible for corporate managers’ self-serving attribution bias and this bias is

associated with financial policies. Therefore, the results lend support to the management

overconfidence literature by strengthening its cognitive foundation.

23

5. Conclusion

I find significant evidence that managers tend to use first-person pronouns in MD&As

more often when firm performance is better. This tendency increases with a firm’s

information environment complexity and past success. Within specific MD&As,

managers tend to use more self-referencing pronouns when the sentence has positive tone.

This evidence suggests that managers have self-serving attribution bias. Management

SAB leads to overconfidence: I find that when managers have more SAB, they are more

likely to include forward-looking statements in the MD&As and issue earnings forecasts;

their forward-looking statements are more homogeneous in tone and their earnings

forecasts tend to be overly optimistic. Also, firms whose managers have SAB tend to

overinvest in that they have higher investment-cash flow sensitivity and more negative

stock returns around merger announcements. Finally, these firms tend to have higher

leverage, use more long-term debt, are more likely to repurchase stock, and are less likely

to issue dividends. Collectively, evidence in this paper supports the existence of

management self-serving attribution bias and suggests that it is important to model this

bias and the resulting management overconfidence in economic analysis.

24

References

Baker, M., Ruback R., Wurgler, J., 2007, Behavioral corporate finance, in Handbook in

Corporate Finance: Empirical Corporate Finance, edited by Espen Eckbo

(Elsevier/North Holland).

Barber, B., Odean, T., 2001, Boys will be Boys: Gender, Overconfidence, and Common

Stock Investment, Quarterly Journal of Economics, February 2001, Vol. 116, No. 1,

261-292.

Barber, B., Odean, T., 2002, Online investors: do the slow die first? Review of Financial

Studies, Vol. 15, No. 2, 455-487.

Ben-David, I., Graham J., Harvey, C., 2007, Managerial overconfidence and corporate

policies, Working paper, Duke University.

Billett, M., Qian, Y., 2008, Are Overconfident CEOs Born or Made? Evidence of Self-

Attribution Bias from Frequent Acquirers, Management Science, Vol. 54, No.6,

1037-1051.

Brown, S.V., and Tucker J., 2010. Large-Sample Evidence on Firms’ Year-Over-Year

MD&A Modifications, Working paper, University of Florida.

Campbell, W. K., Sedikides, C. 1999. Self-threat magnifies the self-serving bias: A meta-

analytic integration. Review of General Psychology, 3, 23–43.

Chatterjee, A. and Hambrick, D. C., 2007. It’s all about me: Narcissistic CEOs and their

effects on company strategy and performance. Administrative Science Quarterly, 52,

351–386.

Choi, D., Dong, L., 2008. A test of the self-serving attribution bias: evidence from mutual

funds, Working paper, Hong Kong University of Science and Technology.

25

Coller, M., and T. L. Yohn. 1997. Management Forecasts and Information Asymmetry:

An Examination of Bid-Ask Spreads. Journal of Accounting Research 35 (2): 181-

191.

Daniel, Kent, Hirshleifer, David, and Subrahmanyam, Avanidhar, 1998, Investor

psychology and security market under- and overreactions, Journal of Finance, LIII

No.6.

De Long, J. Bradford, Andrei Shleifer, Lawrence Summers, and Robert J. Waldmann,

1991, The survival of noise traders in financial markets, Journal of Business 64, 1–20.

Doukas, J., Petmezas, D., 2007, Acquisitions, Overconfident Managers and Self-

attribution Bias, European Financial Management, Vol. 13, No. 3, 2007, 531–577

Fazzari, Steve, R. Glenn Hubbard, and Bruce Peterson, 1988, Financing constraints and

corporate investment, Brookings Papers on Economic Activity, 141–195.

Healy, J.F. 2008. Statistics: A Tool for Social Research. Wadsworth Publishing; 8 edition

Hutton, A. P., and P. C. Stocken. 2007. Effect of Reputation on the Credibility of

Management Forecasts. Working Paper, Boston College and Dartmouth College.

Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–

498.

Larwood, L., Whittaker, W., 1977, Managerial myopia: self-serving biases in

organizational planning, Journal of Applied Psychology, Vol. 62, No.2, 194-198.

Li, Feng, 2010, The information content of forward-looking statements in corporate

filings: A Naïve Bayesian approach, Journal of Accounting Research, Forthcoming.

Malmendier, Ulrike, and Geoffrey Tate, 2005, CEO overconfidence and corporate

investment, Journal of Finance. December, vol. 60 (6), pp. 2661-2700.

26

Malmendier, Ulrike, and Geoffrey Tate, 2008, Who makes acquisitions? CEO

overconfidence and the market’s reaction, Journal of Financial Economics, vol. 89(1),

pp. 20-43.

Malmendier, Ulrike, Geoffrey Tate, and Jonathan Yan, 2010, Managerial beliefs and

corporate financial policies. NBER working paper 15659.

Manning, C., and H. Schutze, 1999. Foundations of Statistical Natural Language

Processing (The MIT Press: Boston, MA).

Miller, D. T., & Ross, M. (1975). Self-serving biases in attribution of causality: Fact or

fiction? Psychological Bulletin, 82, 213-225.

Moeller, Sara B., Frederik P. Schlingemann, and René M. Stulz, 2004, “Firm size and the

gains from acquisitions.” Journal of Financial Economics Vol. 73, No. , pp. 201-228.

Nagar, V., D. Nanda, and P. Wysocki. 2003. Discretionary Disclosure and Stock-based

Incentives. Journal of Accounting and Economics 34: 283-309.

Odean, T., 1998, Volume, volatility, price, and profit when all traders are above average,

Journal of Finance, 53, 1887-1934.

Odean, T., 1999, Do investors trade too much? American Economic Review, 89, 1279-

1298.

Patell, J. M. 1976. Corporate Forecasts of Earnings per Share and Stock Price Behavior:

Empirical Tests. Journal of Accounting Research 14 (2): 246-276.

Pennebaker, J., Francis, M., Booth, R., 2001, Linguistic Inquiry and Word Count 2001

manual, University of Texas.

Penman, S. H. 1980. An Empirical Investigation of the Voluntary Disclosure of

Corporate Earnings Forecasts. Journal of Accounting Research 18 (1): 132-160.

27

Pyszczynski, T., & Greenberg, J. (1987). Toward an integration of cognitive and

motivational perspectives on social inference: A biased hypothesis-testing model.

Advances in Experimental Social Psychology, 20, 297–340.

Russo, J.E., Schoemaker, P, 1992, Managing overconfidence, Sloan Management Review,

Volume 33, No. 2, 7-17.

Schlenker, B. R. 1980. Impression Management: The Self-concept, Social Identity, and

Interpersonal Relations. Monterey, CA: Brooks/Cole.

U.S. Securities and Exchange Commission (SEC), 1980. Securities Act Release No. 6231

(September 2).

Sedikides, C., Campbell W.K., Reeder G., Elliot A., 1998, The self-serving bias in a

relational context, Journal of Personality and Social Psychology. Vol. 74, No 2., 378-

386.

Shepperd, J., Malone, W., Sweeny, K., 2008, Exploring causes of the self-serving bias,

Social and Personality Psychology Compass, 10, 895-908.

Simon, H. A. (1956). Rational choice and the structure of environments. Psychological

Review, 63, 129–138.

Zuckerman, M.1979, Attribution of success and failure revisited, or: The motivational

bias is alive and well in attribution theory. Journal of Personality, 47, 245-28.

28

Table 1 Descriptive statistics

Panel A: Variables for the empirical tests

Variable Obs Mean Std. P25 Median P75 We 38953 1.27 1.73 0.02 0.12 2.78 Other 38953 0.12 0.10 0.06 0.10 0.17 Self 38953 1.15 1.70 -0.06 0.04 2.65 ΔSelf 38953 0.15 0.86 -0.08 0.00 0.11 Ret 38953 0.19 0.97 -0.24 0.06 0.38 Size 38877 5.53 1.99 4.13 5.44 6.85 Ln(assets) 38953 5.83 2.02 4.35 5.77 7.15 Q 38786 1.93 1.63 1.05 1.33 2.09 Firmage 38945 0.15 0.11 0.08 0.12 0.19 RETVOL 37765 0.50 0.12 0.08 0.12 0.19 NFLS 37773 228 187 106 179 290 Ln(NFLS) 37773 5.17 0.73 4.66 5.19 5.67 Forecast 37773 0.11 0.32 - - - FERR 4268 0.04 0.55 -0.24 -0.04 0.29 IQV 37773 0.81 0.14 0.75 0.84 0.91 Leverage 30445 0.24 0.26 0.02 0.19 0.38 LTD 25814 0.72 0.33 0.56 0.87 0.98 Repurchase 30583 0.30 0.46 - - - Dividends 30583 0.32 0.47 - - -

Notes: This table shows the descriptive statistics of the variables examined in the empirical tests. We is the percentage of first-person pronouns in the MD&A. Other is the percentage of second- and third-person pronouns in the MD&A. Self is We minus Other. ΔSelf is the year-to-year change in Self. Ret is the annual stock return during the fiscal year. Size is the logarithm of the market value of equity (Compustat annual item 25 * item 199). Ln(Asstes) is the logarithm of the book value of assets (item 6). Q is the market value (item 6 * item 199 + item 181) divided by the book value of the assets (item 6). Firmage is the average number of years since a firm shows up in CRSP monthly stock return file. RETVOL is the standard deviation of the monthly stock returns during the fiscal year. NFLS is the number of forward-looking statements in MD&As; Ln(NFLS) is the logarithm of NFLS. Forecast is a dummy variable that equals 1 if the manager of a firm makes forecasts of next quarter’s earnings per share after the 10-K filing date and 0 otherwise. FERR is the earnings per share forecast minus the actual earnings per share scaled by the book value of assets per share. IQV is the index of qualitative variation of the tone of the forward-looking statements in MD&A. Leverage is the book value of total debt (long-term plus short-term) divided by the sum of the book value of total debt and the market value of equity at the end of the fiscal year. LTD is the percentage of long-term debt in total debt (item 9 / (item 9 + item 34)). Repurchase is a dummy variable that equals 1 if purchase of common and preferred stock (item 115 / (item 6 – item 181)) is greater than 1% of equity, and zero otherwise. Dividends is a dummy variable that equals 1 if a firm declares dividends in the fiscal year (i.e., item 21 > 0) and zero otherwise.

29

Panel B: Mean values of first-person pronouns and second- and third-person pronouns in typical text domains

Emotion writing Control writing Books Talking We 3.49 3.22 0.49 1.68 Other 1.13 0.50 1.27 1.65

Notes: This table shows the percentages of first-personal pronouns (We) and second- and third-personal pronouns (Other) of the text in the research articles surveyed by Pennebaker et al. (2001, Table 3). Emotion writing studies require participants to write about their emotions and thoughts about personally relevant topics; Control Writing involves writing about non-emotional topics, such as plans for the day or descriptions of ordinary objects or events; Books refers to a semi-random sample of pages from the 30 best-selling fiction books of 1995; Talking files come from transcripts collected from individuals who are talking in non-experimental settings (i.e., correlational studies).

30

Table 2 Regression of managers’ self-referencing tendency on firm performance

Full sample Full sample Sub-sample Sub-sample Sub-sample

(1) (2) (3) (4) (5) Variable ΔSelf ΔSelf ΔSelf ΔSelf ΔSelf Ret 0.11*** 0.09*** 0.14*** 0.25* 0.08** (4.40) (4.98) (3.08) (2.14) (2.36) Q 0.03** 0.02* 0.03 0.00 (2.16) (2.15) (1.34) (0.34) Ln(assets) 0.01* 0.01 0.00 0.01** (1.78) (1.71) (0.42) (2.86) Firmage -0.00 -0.00 -0.00 -0.00 (-0.68) (-1.52) (-0.77) (-0.33) Year fixed effects Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Auditor fixed effects Yes Yes Yes Yes Yes Standard error cluster by

year cluster by

year cluster by

year cluster by

year cluster by

year Observations 35558 35430 18981 5001 9719 R-squared 0.04 0.05 0.03 0.04 0.02

Notes: All variables are as defined in Table 1 Panel A. In coding the auditor fixed effects, all non-big-eight auditors are grouped together. T-statistics is shown in parentheses. The sub-sample in column (3) includes firms whose stock returns are between 25 and 75 percentiles of the returns of firms in the same 2-digit SIC industry. The sub-sample in column (4) includes firms whose stock returns are within the 5 percent range of the median returns of firms in the same 2-digit SIC industry in the same year. The sub-sample in column (5) includes firms whose idiosyncratic stock returns are in the bottom quartile of all firms in my sample.

31

Table 3 Managers’ self-serving attribution bias as a function of firm size, number of business segment, and past success

(1) (2) (3) (4) Variable ΔSelf ΔSelf ΔSelf ΔSelf Ret 0.08*** 0.05 0.00 -0.09** (3.78) (1.23) (0.10) (-2.79) Size 0.01 0.00 (1.33) (0.92) Ret * Size 0.01* 0.02*** (2.06) (3.89) Num_seg -0.02 -0.02 (-0.86) (-0.98) Ret * Num_seg 0.07** 0.06*** (2.86) (3.10) Success -0.02 -0.02 (-1.37) (-1.11) Ret * Success 0.05*** 0.04* (3.52) (1.93) Q 0.02* 0.02* 0.02 0.01 (1.83) (1.96) (1.68) (1.11) Firmage -0.02 0.00 0.00 -0.01 (-0.65) (0.12) (0.16) (-0.59) Year fixed effects Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Auditor fixed effects Yes Yes Yes Yes Standard error cluster by

year cluster by

year cluster by

year cluster by

year Observations 31006 35329 20103 18227 R-squared 0.05 0.05 0.04 0.04

Notes: Size is the logarithm of the market value of equity (Compustat annual item 25 * item 199). Num_seg is the logarithm of the number of business segments from Compustat. Success is a dummy variable that equals 1 if a firm’s stock return is above industry median in the last two years and 0 otherwise. All the other variables are as defined in Table 1 Panel A. T-statistics is shown in parentheses.

32

Table 4 Association between tone and self-referencing tendency of the causation sentences in MD&As

(1) (2) (3) (4) Variable Self Self Self Self Positive_tone 0.32*** 0.37*** 0.33*** 0.37*** (4.78) (4.68) (5.39) (5.18) MDA_length -0.37*** -0.35*** (-10.78) (-14.82) Size 0.18*** 0.20*** (8.51) (6.58) Q -0.02 -0.05* (-1.16) (-1.98) Firmage -0.01 -0.01 (-1.66) (-1.99)* Year fixed effects Yes No Yes No Firm fixed effects Yes No Yes No Auditor fixed effects Yes No Yes No Firm fixed effects * year fixed effects No Yes No Yes Standard error cluster by

year cluster by

year cluster by

year cluster by

year Observations 3165111 3562110 1659908 1855964 R-squared 0.30 0.40 0.26 0.38

Notes: The sample used in the regressions consists of all the sentences that contain at least one “causative” words defined by LIWC, such as “because,” “effect,” and “hence.” In columns (3) and (4), the sample requires a firm to have at least 8 years of MD&A. Positive_tone is a dummy variable that equals one if the probability of a sentence being positive predicted by the Naïve Bayesian algorithm is the highest among the three categories (positive, neutral, and negative), and zero otherwise. MDA_length is the logarithm of the number of words in an MDS&A. All the other variables are as defined in Table 1 Panel A. T-statistics is shown in parentheses.

33

Table 5 Managers’ self-serving attribution bias and managerial forward-looking statements and earnings forecasts

(1) (2) (3) (4) (5) (6) (7) (8) Variable Ln(NFLS) Ln(NFLS) Forecast Forecast FERR FERR IQV IQV SAB 0.049*** 0.025*** 0.103*** 0.101** 0.052*** 0.038* -0.003* -0.007*** (7.91) (4.42) (2.87) (2.07) (3.15) (1.73) (-1.92) (-4.32) Size 0.130*** 0.017** 0.398*** 0.501*** 0.022*** -0.017 -0.007*** -0.009*** (68.36) (2.51) (38.58) (8.66) (3.42) (-0.58) (-16.61) (-5.52) Q -0.016*** -0.002 -0.016** 0.010 0.010*** 0.013** -0.002*** -0.000 (-9.10) (-1.42) (-2.38) (1.00) (4.27) (1.98) (-5.09) (-0.67) Firmage 0.001*** -0.006 -0.011*** 0.040** -0.001** -0.026*** 0.000*** -0.000 (2.94) (-1.58) (-8.22) (2.21) (-2.53) (-2.59) (7.17) (-0.07) Earn -0.230*** -0.149*** 1.223*** 0.829** -0.258*** 0.310* 0.021*** 0.001 (-16.11) (-5.54) (11.03) (2.28) (-2.95) (1.77) (6.44) (0.25) RETVOL 0.466*** 0.133*** -0.267 -2.047*** -0.042 -0.283 -0.079*** -0.034*** (12.30) (3.30) (-1.40) (-4.33) (-0.29) (-1.25) (-9.57) (-3.34) Year effects Yes Yes Yes Yes Yes Yes Yes Yes Industry effects Yes No Yes No Yes No Yes No Firm effects No Yes No Yes No Yes No Yes Standard error robust cluster by

firm robust cluster by

firm robust cluster by

firm robust cluster by

firm Observations 37389 25367 37332 10015 4227 4227 37389 25367 R-squared 0.36 0.71 - - 0.09 0.55 0.07 0.45

Notes: SAB is a dummy variable that equals 1 if itε is above the median and 0 otherwise, where itε is the residual from the pooled regression of Self on Ret. All other variables are as defined in the Notes to Table 1 Panel A. T-statistics is shown in parentheses for columns (1), (2), (5), (6), (7) and (8). Z-statistics is shown in parentheses for columns (3) and (4).

34

Table 6 Managers’ self-serving attribution bias and corporate investment-cash flow sensitivity

(1) (2) (3) (4) (5) (6) Variable CAPX CAPX CAPX CAPX CAPX CAPX CF -0.174** -0.099 -0.132 -0.033 -0.033 0.172 (-2.12) (-0.40) (-0.46) (-0.00) (-0.00) (0.58) Q 0.118*** 0.115*** 0.116*** 0.112*** 0.112*** 0.128** (3.88) (3.40) (3.42) (3.63) (3.30) (2.17) CF * Q -0.018 -0.018 -0.022 -0.025 -0.025 -0.020 (-1.44) (-1.43) (-1.52) (-1.52) (-1.32) (-0.96) SAB -0.236** -0.224** -0.221** -0.202** -0.202* -0.122 (-2.24) (-2.13) (-2.12) (-2.10) (-1.73) (-1.60) SAB * CF 0.223** 0.225** 0.224** 0.225** 0.225* 0.148* (2.08) (2.01) (2.06) (2.11) (1.72) (1.70) Size -0.070 -0.076 -0.069 -0.069 0.004 (-1.45) (-1.47) (-1.46) (-1.08) (0.03) Size * CF 0.016 0.030 0.048 0.048 -0.019 (0.47) (0.70) (1.00) (0.73) (-0.91) Earn 0.739* 0.690* 0.442 0.442 1.142 (1.78) (1.72) (1.16) (1.03) (1.32) Earn * CF 0.145 0.128 0.016 0.016 0.042 (1.33) (1.25) (0.13) (0.12) (0.27) Firmage -0.003 -0.004* -0.003 -0.003 0.026 (-1.45) (-1.91) (-1.36) (-1.36) (0.82) Firmage * CF 0.004 0.005 0.002 0.002 0.006 (1.11) (1.32) (0.66) (0.60) (1.03) RETVOL 1.397*** 1.361*** 0.975*** 0.975*** 1.268 (2.72) (2.75) (2.83) (2.88) (1.64) RETVOL * CF -0.858 -1.020* -1.040* -1.040 -1.213 (-1.54) (-1.69) (-1.71) (-1.59) (-1.43) Year effects Yes Yes Yes Yes Yes Yes Industry effects Yes Yes Yes Yes Yes No Year effects * CF No No Yes Yes Yes No Industry effects * CF No No No Yes Yes No Firm effects No No No No No Yes Standard error robust robust robust robust cluster by firm cluster by firm Observations 28740 28711 28711 28711 28711 19580 R-squared 0.01 0.01 0.01 0.03 0.03 0.17

Notes: The dependent variable is CAPX, the capital expenditure (Compustat annual item 128) divided by the amount of property, plant, and equipment at the beginning of the year (item 8). CF is earnings before extraordinary items (item 18) plus depreciation (item 14) and is normalized by the amount of property, plant, and equipments at the beginning of the year (item 8). SAB is a dummy variable that equals 1 if itε is above the median and 0 otherwise, where itε is the residual from the pooled regression of Self on Ret.All other variables are as defined in Table 1 Panel A. T-statistics is shown in parentheses. .

35

Table 7 Managers’ self-serving attribution bias and acquisition announcement returns

(1) (2) (3) Variable CAR[-1,1] CAR[-1,1] CAR[-1,1] SAB -0.006*** -0.006** -0.005** (-3.42) (-2.50) (-2.00) Size -0.002** 0.000 (-2.54) (0.14) Q -0.001 0.000 (-1.25) (0.01) Firmage 0.000 0.000 (1.12) (0.32) Earn 0.018* 0.011 (1.80) (0.54) RETVOL 0.064** 0.041 (2.14) (1.27) Cash 0.005** 0.004 (2.13) (1.08) Deal_size -0.002* -0.003*** (-1.95) (-2.93) Diversify -0.000 -0.006 (-0.11) (-1.42) Year fixed effects Yes Yes Yes Industry fixed effects Yes Yes No Firm fixed effects No No Yes Standard error robust cluster

by firm cluster by firm

Observations 7712 7615 7615 R-squared 0.03 0.04 0.64

Notes: The dependent variable is CAR[-1,1], the cumulative abnormal return measured over the window [-1,1] where day 0 is the deal announcement date. The daily abnormal return is calculated as raw returns subtract the CRSP value-weighted daily index returns. SAB is a dummy variable that equals 1 if itε is above the median and 0 otherwise, where itε is the residual from the pooled regression of Self on Ret. Cash is the percentage of cash payment in the deal. Deal_size is the log of the value of the transaction scaled by the market value of equity of the acquirer. Diversify is a dummy variable that equals 1 if the target and the acquirer do not have the same 2-digit SIC code and 0 otherwise. All other variables are as defined in Table 1 Panel A. T-statistics is shown in parentheses

36

Table 8 Managers’ self-serving attribution bias and corporate financing policies

(1) (2) (3) (4) (5) (6) (7) Variable Leverage Leverage LTD LTD Repurchase Repurchase Dividends SAB 0.038*** 0.015*** 0.012*** 0.007* 0.069** 0.082** -0.076** (15.68) (9.23) (2.90) (1.86) (2.47) (2.22) (-2.24) Size -0.027*** -0.093*** 0.034*** 0.025*** 0.169*** -0.119*** 0.339*** (-29.17) (-32.78) (25.40) (5.54) (18.44) (-2.85) (12.89) Q -0.016*** 0.002*** -0.011*** -0.002 0.019** -0.002 -0.272*** (-12.33) (2.96) (-5.66) (-1.05) (2.45) (-0.15) (-6.63) Firmage 0.001*** 0.001 -0.001*** 0.002 0.008*** -0.023 0.046*** (9.66) (0.27) (-6.03) (1.17) (7.73) (-1.23) (11.49) Earn 0.036*** 0.027*** 0.096*** 0.039 0.395*** -0.169 4.566*** (6.90) (3.21) (8.96) (1.41) (3.86) (-1.13) (11.43) RETVOL 0.020 0.039*** -0.212*** -0.069** -2.546*** -1.610*** -12.000*** (1.34) (2.92) (-8.04) (-2.17) (-10.30) (-4.73) (-16.65) PPE(-1) 0.000*** 0.000*** -0.000*** 0.000 0.000*** 0.000 0.000 (8.63) (3.66) (-15.21) (0.00) (2.90) (0.22) (0.89) RET(-1) -0.005*** -0.006*** 0.003 -0.000 -0.146*** -0.064*** -0.165*** (-3.35) (-5.07) (1.09) (-0.17) (-6.55) (-2.92) (-5.65) Year effects Yes Yes Yes Yes Yes Yes Yes Industry effects Yes No Yes No Yes No Yes Firm effects No Yes No Yes No Yes No Standard error robust cluster by firm robust cluster by firm robust cluster by firm cluster by firmObservations 27310 19314 23077 16415 27375 27375 27347 R-squared 0.32 0.83 0.16 0.58 - - -

Notes: SAB is a dummy variable that equals 1 if itε is above the median and 0 otherwise, where itε is the residual from the pooled regression of Self on Ret. PPE(-1) is the amount of property, plant, and equipments at the end of last year. RET(-1) is the stock returns from last fiscal year. All other variables are as defined in the Notes to Table 1 Panel A. T-statistics is shown in parentheses for columns (1) to (4). Z-statistics is shown in parentheses for columns (5) to (7).